A hybrid model-based method for leak detection in large scale water distribution networks

During the past decades, the problem of finding leaks in Water Distribution Networks (WDN) has been challenging. The quicker detection of leaks, on one hand prevents water loss and helps avoiding economic and environmental leakage consequences. On the other hand, increasing the speed of leak detection increases the false leak detection that imposes high costs. In this paper, we propose a fast hybrid method using AI algorithms and hydraulic relations for detecting and locating leaks and identifying the volume of losses material. The proposed method relies on simple and cost-effective flow sensors installed on each junction in the pipeline network. We demonstrate how influential features for leak detection would be generated by using hydraulic equations like Hazen-Williams, Darcy-Weisbach and pressure drop. Through exploiting Decision Tree, KNN, random forest, and Bayesian network we build predictive models and based on the pipeline topology, we locate leaks and their pressure. Comparing the results of applying the proposed method on various leak scenarios shows that the proposed method in this paper, outperforms other existing methods.